Specification · The Layer 3 Standard
brand.context

A machine-readable standard for the decision moment.

An open standard for brands to declare their decision-stage evidence in a form AI systems can actually use when they make a recommendation — the concrete implementation of Layer 3 Activation, built to close the Linkage Gap.

By Tim de Rosen & Paul Sheals AIVO Standard · v2.0 Published July 2026 ≈ 7 min read

An AI agent evaluating a purchase on a consumer’s behalf doesn’t browse your website and form a considered view the way a person would. It queries structured knowledge, retrieves context where it exists, weighs a brand against a set of criteria, and produces a recommendation — often in a single turn.

Until now, no standard existed for a brand to declare its decision-stage evidence in a format built for that process rather than for a human reader or a search crawler. That’s the gap the Linkage Gap research exposed: across 1,427 probes, 87.3% of brands present when a conversation began were displaced before the recommendation — usually because the model had the relevant facts, but didn’t carry them to the moment the decision was made. brand.context is the standard built to close that.

What it is

One file, published where agents look

brand.context is an open, machine-readable JSON-LD file a brand publishes at a predictable path on its own domain:

https://yourbrand.com/.well-known/brand.context

It declares the specific, structured evidence that determines whether a brand survives to an AI’s final recommendation — the right facts, in the right form, ready to be surfaced at the decision turn. It’s a formal specification: it uses RFC 2119 conformance language (MUST / SHOULD / MAY), commits to a six-month deprecation window on breaking changes, and grounds every claim in published, DOI-cited research.

Built for trust

You cite — you don’t just claim

What separates brand.context from a marketing file is that it’s built to be believed by a machine that has learned to discount hype. Two fields do that work.

evidence_status

Verified vs self-declared

Every claim is tagged by how checkable it is. A consuming AI should weight the two very differently.

verified — backed by a source a third party can check: a clinical study, a regulatory record, a dated product page.
self-declared — the brand’s own honest assessment, with no independent corroboration.
gap_type

What evidence can, and can’t, fix

Every field is honest about what publishing it will achieve — a rare thing for a brand standard to admit.

linkage — a fact that empirically changes a model’s recommendation when it’s present at the decision turn.
reasoning_risk — a structural fit judgement that evidence alone will not override.
What’s inside

Evidence, by category

Beyond a brand-identity layer (so the model links to the right entity) and a competitive-positioning layer, the core of the schema is eight evidence categories — a brand fills in the ones that apply to its sector.

01

Clinical evidence

Active ingredients and cited clinical backing — skincare, supplements, OTC health.

02

Lifestyle fit

Which purchase criteria you lead on — banking, automotive, subscriptions.

03

Regulatory access

Which formulation is available, and where — OTC and nutraceuticals.

04

Technology currency

Your core technology vs a rival’s “newer-science” framing — electronics, SaaS, haircare.

05

Product version

Your current model vs your own superseded versions still in training data.

06

Contextual relevance

Where you genuinely win vs your general-category standing (a reasoning-risk field).

07

Availability

Where you actually ship, sell and fulfil — regional and expanding brands.

08

Historical narrative

Current, verifiable status vs an outdated decline event in the public record.

Where it fits

Not another sitemap

A whole machine-readable layer is forming for AI — schema.org, LLMs.txt, MCP, Google’s Open Knowledge Format and Agentic Resource Discovery, EntityMap. Those address discovery and packaging: which pages matter, what a domain can do, how knowledge is bundled for retrieval. None of them take a position on which facts about a brand determine a purchase outcome.

“brand.context occupies a narrower layer than any of them: the specific, tested evidence that decides whether a brand survives an AI’s final recommendation.”

It’s complementary, not competitive — a brand can publish an entity index for general knowledge and a brand.context file for the evidence that decides the sale.

Stated plainly

Honest about where it stands

The standard is candid about its own maturity, because a standard that overclaims can’t be trusted. There is no guaranteed consumption mechanism today — no major AI platform yet crawls for brand.context files by name. (This isn’t unique to it: independent data found 97% of published LLMs.txt files received zero requests.)

So why publish now? The present value is real but partial: retrieval-augmented systems that index structured web content can already pick it up, and it’s a forward position for the context-graph infrastructure now being built. It’s an early investment with an immediate-but-partial return — and the standard says so, in every field where the distinction applies.

The full specification

Read the complete standard

This article is an overview. The full v2.0 specification defines the conformance rules, the complete field-by-field schema, the evidence-confidence model, the relationship to other standards and the honest limitations — published open-access on Zenodo with a permanent DOI.

Citation: de Rosen, T. & Sheals, P. (2026). brand.context v2.0: A Machine-Readable Evidence Standard for Closing the Linkage Gap in Agentic Commerce. AIVO Standard. Zenodo. https://doi.org/10.5281/zenodo.21262005 · Licensed CC-BY-4.0.